US 12,112,260 B2
Metrology apparatus and method for determining a characteristic of one or more structures on a substrate
Lorenzo Tripodi, Eindhoven (NL); Patrick Warnaar, Tilburg (NL); Grzegorz Grzela, Eindhoven (NL); Mohammadreza Hajiahmadi, Rotterdam (NL); Farzad Farhadzadeh, Eindhoven (NL); Patricius Aloysius Jacobus Tinnemans, Hapert (NL); Scott Anderson Middlebrooks, Duizel (NL); Adrianus Cornelis Matheus Koopman, Hilversum (NL); Frank Staals, Eindhoven (NL); Brennan Peterson, Longmont, CO (US); and Anton Bernhard Van Oosten, Lommel (BE)
Assigned to ASML Netherlands B.V., Veldhoven (NL)
Filed by ASML Netherlands B.V., Veldhoven (NL)
Filed on May 29, 2019, as Appl. No. 16/424,811.
Claims priority of application No. 18176718 (EP), filed on Jun. 8, 2018; application No. 18190559 (EP), filed on Aug. 23, 2018; and application No. 18206279 (EP), filed on Nov. 14, 2018.
Prior Publication US 2019/0378012 A1, Dec. 12, 2019
Int. Cl. G06N 3/08 (2023.01); G01B 11/02 (2006.01); G01N 21/55 (2014.01); G06T 7/00 (2017.01)
CPC G06N 3/08 (2013.01) [G01B 11/02 (2013.01); G01N 21/55 (2013.01); G06T 7/0006 (2013.01); G06T 7/001 (2013.01); G01B 2210/56 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/30148 (2013.01)] 19 Claims
OG exemplary drawing
 
1. A method comprising:
obtaining a low-quality input image of a metrology target formed on a substrate by a lithographic process, wherein the metrology target comprises at least one feature corresponding to a characteristic of interest, and wherein the metrology target comprises at least two sub-targets having periodic patterns arranged orthogonal to each other;
performing a calibration phase to train a neural network, the calibration phase comprising:
obtaining first low-quality training images of a calibration metrology target formed on a calibration substrate by a calibration lithographic process having different values of the characteristic of interest and corresponding known values of the characteristic of interest, the metrology target and the calibration metrology target comprising similar structures; and
using the first low-quality training images and corresponding known values of the characteristic of interest to train the neural network to infer values of the characteristic of interest from the first low-quality training images,
wherein the corresponding known values of the characteristic of interest are obtained from second high-quality training images which are a higher quality than the first low-quality training images, the second high-quality training images being of a corresponding calibration metrology target as the first low-quality training images;
using the trained neural network to determine the characteristic of interest relating to the metrology target based on the low-quality input image; and
using the trained neural network to determine one or more optimal measurement recipes, wherein each measurement recipe comprises a combination of measurement settings,
wherein the characteristic of interest comprises overlay, focus, an energetic illumination characteristic, a geometric non-telecentricity illumination characteristic, dose, or an aberration, and
wherein training of the neural network is limited to the metrology target.